Install
openclaw skills install @mlwood-dev/everwood-aidlcStrict human-gated AIDLC planning for non-trivial software work, with per-gate deconfliction review.
openclaw skills install @mlwood-dev/everwood-aidlcStrict AI-Driven Development Life Cycle for non-trivial software work.
Human-gated Inception (Gates 0–4) before Construction. Gate Deconfliction runs before each human present. Workspace scratch under aidlc-sessions/ is the sole content source of truth shipped with this skill.
This skill expects the agent to:
python3 / bash helpers shipped under {baseDir}/scripts/{workspace}/aidlc-sessions/** (gates, approvals, meta, deconfliction reports)It does not require Redis, a Cache UI, Docker, network services, or external publish steps for normal use.
Starting AIDLC creates/updates local files:
{workspace}/aidlc-sessions/<uuid>/
session-id
meta.json
APPROVALS.md
gates/*.md
gates/*.deconfliction.md
Tell the user once at session init that planning state will be written there.
Do not put secrets, credentials, tokens, or sensitive customer data in gate artifacts.
Activate for:
/aidlc, start AIDLC, Using AI-DLCDo not auto-activate on vague “maybe refactor” chat. If scope is unclear, ask whether to run AIDLC before writing session files.
Do not force full AIDLC for trivial one-liners unless the user asks.
Resolve {baseDir} as this skill directory. Resolve {workspace} as the active OpenClaw workspace root.
| Asset | Path |
|---|---|
| Core workflow | {baseDir}/references/core-workflow.md |
| Gate templates | {baseDir}/templates/gate-*.md |
| Deconfliction template | {baseDir}/templates/gate-deconfliction.md |
| Session init | {baseDir}/scripts/session-init.py |
| Gate lock (scratch SoT) | {baseDir}/scripts/gate-lock.py |
Workspace scratch (content SoT):
{workspace}/aidlc-sessions/
CURRENT
<uuid>/
session-id
meta.json
APPROVALS.md
gates/
<gate>.md
<gate>.deconfliction.md
{baseDir}/references/core-workflow.md and follow it (includes Gate Deconfliction).aidlc-sessions/CURRENT or history), offer resume from the last approved gate.python3 "{baseDir}/scripts/session-init.py" --root "{workspace}" --objective "<intent>" --json
gates/ directory.Complete gates in order. For each gate:
{baseDir}/templates/).Goal: Catch contradictions, open questions, missing constraints, and weak option tradeoffs before approval.
Default mechanism: spawn an isolated reviewer subagent, then wait for its report.
sessions_spawn(
taskName="aidlc-deconflict-<gate-id>",
mode="run",
context="isolated",
task="AIDLC Gate Deconfliction reviewer (read-only). Gate=<id>, session=<uuid>, objective=<…>.
Review current draft vs prior locked gates. Do not implement or approve.
Structured report per templates/gate-deconfliction.md:
contradictions, open questions, missing constraints, option tradeoffs,
cross-gate drift, suggested revisions, residual risks, verdict clean|issues-found.
Draft: <paste or path>"
)
→ sessions_yield until complete
→ write report to aidlc-sessions/<uuid>/gates/<gate>.deconfliction.md
→ merge fixes into gate draft; re-run if material changes
→ present gate + summary to human
Use template {baseDir}/templates/gate-deconfliction.md.
Rules:
DECONFLICTION (self-review) — do not skip silently.Intent & success criteria; greenfield vs brownfield; constraints; existing assets; open questions.
Complexity (Low / Medium / High); risks; dependencies; recommended depth: full / adaptive / minimal.
Units of Work with dependencies; suggested owner: OpenClaw subagent / human.
Key architectural/design choices with rationale. Deconfliction pays extra attention to option tradeoffs.
Ordered steps, parallelism, agent usage, verification/acceptance criteria. Deconfliction checks plan vs locked decisions.
When the user explicitly approves (Approve and Continue or equivalent):
gates/ dir (keep latest .deconfliction.md beside it).python3 "{baseDir}/scripts/gate-lock.py" \
--root "{workspace}" \
--gate gate-N-... \
--artifact-file "{workspace}/aidlc-sessions/<uuid>/gates/<gate>.md" \
--status approved \
--objective "<intent>"
gate-lock.py updates meta.json and appends APPROVALS.md.Only after Gate 4 is approved/locked:
If a new material decision appears, stop and re-enter the appropriate gate (with deconfliction) rather than deciding silently.
Full depth for medium/high complexity or ambiguous requests. Collapse remaining gates only when the user explicitly wants a lightweight path or Gate 1 recommends minimal and the user approves that recommendation. Lightweight path may waive deconfliction only if the user says so explicitly.
If aidlc-sessions/CURRENT or conversation history shows a prior run:
meta.json + gates/ + APPROVALS.md (+ any *.deconfliction.md)Registry slug: everwood-aidlc
(openclaw-* / *-openclaw slugs are reserved — do not use openclaw-aidlc.)
clawhub install everwood-aidlc
# or
openclaw skills install everwood-aidlc
openclaw skills install everwood-aidlc --global
openclaw skills install /path/to/openclaw-aidlc --force
openclaw skills install git:https://github.com/Everwood-Technologies/openclaw-aidlc.git --force
openclaw skills install /path/to/openclaw-aidlc --global --force
npm i -g clawhub
clawhub login
clawhub skill publish /path/to/openclaw-aidlc \
--slug everwood-aidlc \
--name "Everwood AIDLC (OpenClaw)" \
--version 1.2.0 \
--changelog "remove Redis/Cache UI; scratch-only SoT + Gate Deconfliction" \
--source-repo https://github.com/Everwood-Technologies/openclaw-aidlc \
--no-input
One publish at a time (avoid parallel runs / stale upload tickets).
references/core-workflow.md